Pde Fem

Scientific Computing I
Module 7: An Introduction to Finite Element Methods
Michael Bader
Lehrstuhl Informatik V
Winter 2021/2022
Part I: Introduction to Finite Element Methods
The Model Problem
FEM Main Ingredients

Weak Forms and Weak Solutions
Test and Shape Functions
Assembling the System of Equations
Example Problem: 1D Poisson

Example Problem, Weak Form and Nodal Basis
Stiffness Matrix for Linear Nodal Basis Functions
A Road to Theory
Time-Dependent Problems
Michael Bader | Scientific Computing I | Introduction to Finite Element Methods | Winter 2021/2022 2
The Model Problem: Poisson Equation
• 2D Poisson Equation on unit square:
∂2 ∂2
− 2
u(x, y ) − u(x, y) = f (x, y ) in Ω = (0, 1)2
∂x ∂y 2
• Dirichlet boundary conditions:
u(x, y) = g(x, y ) on ∂Ω
• for general operator L instead of ∆: similar concept

• new for FEM: compute a function to approximate u(x, y )
(instead of approximate values at certain grid points or in grid cells)
Finite Elements – Main Ingredients
1. solve weak form of PDE to reduce regularity properties
Z Z
00 00
−u = f −→ − vu dx = vf dx for “all” v
Ω Ω
→ allows additional weak solutions (esp. after integration by parts)

2. compute a function as numerical solution
→ search in a function space Wh :
X
uh = uj ϕj (x), span{ϕ1 , . . . , ϕJ } = Wh
j
3. find weak solutions of simple form:

for example piecewise linear functions and choose basis functions with
local support (“hat functions”)
→ leads to system of linear equations
Weak Form and Weak Solutions
Motivation:
• an approximate solution (i.e., function) will likely not satisfy the PDE
• we thus need to “weaken” the requirements on the approximating function
The weak form is obtained via 3 steps: −u 00 = f

1. multiply with a test function v ↓
vu 00 dx
R R
(from a certain function space V ) − Ω
= Ω
vf dx for all v
2. integrate over the domain

⇒ integral over function product equiv. to scalar product on functions
3. (optional) perform “integration by parts”
⇒ “transfers” derivatives from solution u to test functions v
⇒ reduces smoothness requirements on u
⇒ “real solution” u also solves the weak form

however, additional, weak solutions are now allowed
Example: Weak Form of the Poisson Equation
• Poisson equation with Dirichlet conditions:
−∆u = f in Ω, u = 0 on δΩ
• weak form: Z Z
− ∆u · v dΩ = f · v dΩ ∀v ∈ V
Ω Ω
• apply divergence theorem:

Z Z Z
− ∆u · v dΩ = ∇u · ∇v dΩ − ∇u · v ds
Ω Ω ∂Ω
• choose functions v such that v = 0 on ∂Ω:

Z Z
∇u · ∇v dΩ = f · v dΩ ∀v ∈ V
Ω Ω
Example: Weak Form of the Poisson Equation (2)
• Poisson equation with Dirichlet conditions:
−∆u = f in Ω, u = 0 on δΩ
• transformed into weak form:

Z Z
∇u · ∇v dΩ = f · v dΩ ∀v ∈ V
Ω Ω
• weaker requirements for a solution u:

twice differentiabale → first derivative integrable
• h∇u, ∇v i a bilinear form (i.e. depends on 2 functions u and v );
often written as:
a (u, v ) = hf , v i ∀v ∈ V
A Bit More Motivation for the Weak Form
FEM and the Residual Equation
!
• for the PDE −∆u = f , the residual is defined as r (x) := ∆u(x) + f (x) = 0
• in the weak form (multiply with test function an integrate) we obtain
Z Z

v (x) r (x) = v (x) ∆u(x) + f (x) dx = 0 for all v
Z
• remember that f (x)g(x) dx is the standard scalar product on functions
• hence: the residual of the weak solution has to be orthogonal to all test functions
(and recall all connections between “orthogonal” and “shortest distance”)
FEM vs. Linear Algebra

• consider the system of linear equations Ax = b
• now imagine a “weak form” with scalar product: demand v T Ax = v T b for all v
• if v = eT = (0, . . . 0, 1, 0 . . . , 0), then eT Ax = eT b tests a single equation (i-th row)
i i i
• if you use all unit vectors ei as test functions, only an exact x is also a weak solution
• you may use fewer test functions (and maybe different ones) to allow “weak” solutions
• for the PDE weak form, you cannot use all test functions: these would be infinitely many
Test and Shape Functions
• search for solution functions uh of the form
X
uh = uj ϕj (x)
j
the ϕj (x) are typically called shape or ansatz functions

• the basis functions ϕj (x) build a vector space
(i.e., a function space) Wh
span{ϕ1 , . . . , ϕJ } = Wh
• insert into weak formulation

Z X Z
∇ uj ϕj (x) · ∇v dx = f · v dx ∀v ∈ V
j
Test and Shape Functions (2)
• choose a basis {ψi } of a test space Vh ⊂ V

• then: if all basis functions ψi satisfy
Z X Z
∇ uj ϕj (x) · ∇ψi dx = f · ψi dx ∀ψi
j
P
then all v = i vi ψi (x) ∈ Vh satisfy the equation
• the {ψi } are therefore often called test functions
• we obtain a system of equations for unknowns uj :
one equation for each test function ψi
• Vh is often chosen to be identical to Wh (Ritz-Galerkin method)
→ we then have as many equations as unknowns
Example: Nodal Basis
1
 1
h (x − xi−1 ) xi−1 < x < xi 0,8

 0,6
1
h (xi+1 − x)
ϕi (x) := xi < x < xi+1 0,4

0 otherwise
 0,2
0
0 0,2 0,4 0,6 0,8 1
x
f(x)
3 u(x) 3 u(x)=iii(x)
2 2
1 1
0 0
0 xi 1 0 1
h3=2-3
Assembling the System of Equations
• start from system of equations
Z X Z
∇ uj ϕj (x) · ∇ψi dx = f · ψi dx ∀ψi
j
• ∇ and
R
are linear operators ⇒ rewrite into canonical form:
XZ Z
∇ϕj (x) · ∇ψi (x) dx uj = f · ψi dx ∀ψi
j | {z } | {z }
=:Aij =:fi
• aim: make matrix A sparse → most Aij = 0

• approach: use local basis functions on a discretisation grid
• ψj , ϕj zero everywhere except in grid cells adjacent to grid point xj
• Aij = 0, if ψi and ϕj don’t overlap
Example Problem: 1D Poisson
• in 1D: −u 00 (x) = f (x) on Ω = (0, 1),
hom. Dirichlet boundary cond.: u(0) = u(1) = 0
• weak form:
Z 1 Z 1
0 0
u (x) · v (x) dx = f (x) · v (x) dx ∀v ∈ V
0 0
• computational grid:
xi = ih, (for i = 1, . . . , n − 1); mesh size h = 1/n
• Vh = Wh : piecewise linear functions (on intervals [xi , xi+1 ])
• insert uh = j uj ϕj (x) into weak form:
P
Z 1X Z 1
0 0
uj ϕj (x) · ϕi (x) dx = f (x) · ϕi (x) dx ∀i = 1, . . . , n − 1
0 j 0
| {z }
= j uj 01 ϕ0j (x)·ϕ0i (x) dx
P R
Nodal Basis
 1

 h (x − xi−1 ) xi−1 < x < xi
1
h (xi+1 − x)
ϕi (x) := xi < x < xi+1

0 otherwise

0,8
0,6
0,4
0,2
0
0 0,2 0,4 0,6 0,8 1
x
Nodal Basis – Derivatives and Integrals
Compute derivative of basis functions:
 1
 h
 xi−1 < x < xi
0
ϕi (x) := − h1 xi < x < xi+1

0 otherwise

Compute integrals of weak form:

Z Z xi Z xi+1
1 1 1 1 2
• ϕ0i (x)ϕ0i (x) dx = · dx + − − dx =
xi−1 h h xi h h h
Z Z xi+1
1 1 1
• ϕ0i (x)ϕ0i+1 (x) dx = − dx = −
xi h h h
Z
0 0 1
• ϕi (x)ϕi−1 (x) dx = − (same computation)
h
Z
• ϕ0i (x)ϕ0j (x) dx = 0 for all j 6= i, i ± 1
(supports of basis functions do not overlap!)
Resulting System of Equations
• resulting system matrix A (often called “stiffness” matrix):
 
2 −1
 .. 
1 −1 2 . 

h .. .. 
. . −1 
−1 2
• right hand sides (assume f (x) = α ∈ R):

Z 1 Z 1
f (x) · ϕi (x) dx = α · ϕi (x) dx = αh
0 0
• system of equations very similar to finite differences

• note: for other operators L: same approach!
• exercise: compute right-hand-side vector, if f (x) := fj φj (x)
P
Outlook: A Road to Theory
• requires concepts for convergence of a sequence of functions

(that better and better approximate the solution)
• weak formulation is equivalent to variational approach:
solution u minimises an energy functional
• best approximation property:

u − uhFE ≤ inf ku − v ka

a v ∈Vh
in terms of the norm induced by the bilinear form a (energy norm)

• thus: error bounded by interpolation error
(in energy norm)
(details in lecture Numerical Programming II, e.g.)
Time-Dependent Problems
Example: 1D Heat Equation

• ut = uxx + f on domain Ω = [0, 1] for t ∈ [0, tend ]
• formulate weak form
Z Z Z
ut · v dx = uxx · v dx + f · v dx
Z Z Z
∂
∂t u · v dx = uxx · v dx + f · v dx
• apply discretisation in space using finite element method:

∂
∂t (Mh uh ) = Ah uh + fh
Mh : mass matrix, Ah : stiffness matrix, uh = uh (t)
Time-Dependent Problems (2)
Solve a system of ordinary differential equations:
• after spatial discretisation (Mh constant):
∂
∂t Mh (uh ) = Ah uh + fh
• uh a vector of time-dependent coefficient functions:
T
uh = (u1 (t), . . . , ui (t), . . . , un (t))

• e.g., with forward Euler: Mh u(n+1)
h − uh
(n) (n)
= Ah uh + fh
Outlook: how to deal with inversion of mass matrix?

• approximate Mh by a simpler matrix (diagonal matrix, e.g.)
→ mass lumping
• choose orthogonal basis functions → mass matrix diagonal
• choose entirely element-local basis functions (“discontinuous
Galerkin”) → mass matrix block-diagonal
Part II: Implementation of Finite Element Methods
Element Stiffness Matrices

Element-Oriented Computation of Stiffness Matrices
Example: 1D Poisson
Example: 2D Poisson
Typical Workflow
Reference Elements and Stiffness Matrices

Element-Oriented Computation on Unstructured Meshes
Accumulation of Global Stiffness Matrix
Simple Example: 1D Poisson
Outlook: Extension to 2D and 3D
Outlook: Further Components and Aspects of FEM
Element Stiffness Matrices
Finite Element methods are often used for complicated meshes:
Questions:
• how to set up the system of equations efficiently?
• is it possible to use stencil notation??
⇒ Switch to element stiffness matrices!
Element Stiffness Matrices (2)
• for discretisation → domain Ω subdivided into finite elements Ω(k) :
Ω = Ω(1) ∪ Ω(2) ∪ · · · ∪ Ω(n)
• observation: basis functions are defined element-wisely

Zb Zc Zb
• for computation of integrals, use: f (x) dx = f (x) dx + f (x) dx
a a c
• element-wise evaluation of the integrals:
Z XZ
∇u · ∇v dx = ∇u · ∇v dx
Ω k Ω(k )
Z XZ
f · v dx = f · v dx
Ω k Ω(k )
• leads to local stiffness matrices A(k ) for each element:
Z XZ
∇φj · ∇φi dx = ∇φj · ∇φi dx
Ω k
Ω(k )
| {z } | {z }
=:Aij =:Aij
(k)
• and respective element systems:

Z
(k)
A(k) x = b(k) with bi := f · φi dx
Ω(k )
• accumulate to obtain global system:

X X
A(k) x = b(k )
k k
| {z }
=:A
Some comments on notation:

• assume: 1D problem, n elements (i.e. intervals), linear nodal basis
• in each element only two basis functions are non-zero!
• hence, almost all A(k)
ij are zero:
Z
(k)
Aij = ∇φj · ∇φi dx
Ω(k)
• only 2 × 2 elements of A(k ) are non-zero

• therefore convention to omit a-priori zero columns/rows
⇒ leaves only unknowns that are in Ω(k)
b (k ) (element stiffness matrix)
⇒ notation: write this matrix as A
Example: 1D Poisson
• Ω = [0, 1] splitted into Ω(k) = [xk , xk +1 ]
• nodal basis; leads to element stiffness matrix:

b (k) = 1 1 −1
A h −1 1
• consider case with only three unknowns (i.e., 3 basis functions):

   
1 −1 0 0 0 0
A(1) + A(2) = h1  −1 1 0  + h1  0 1 −1 
0 0 0 0 −1 1
• in stencil notation:
1 1 1
h [ −1 1 0] + h[0 1 − 1] → h [ −1 2 − 1]
Example: 2D Poisson
• −∆u = f on domain Ω = [0, 1]2
• splitted into Ω(i,j) = [xi , xi+1 ] × [xj , xj+1 ]
• bilinear basis functions
ϕij (x, y ) = ϕi (x)ϕj (y )
• “pagoda” functions
Example: 2D Poisson (2)
• leads to element stiffness matrix:
2 − 12 − 12 −1
 
 −1 2 −1 − 12 
Ab (k) = 
 21

− 12 

 −2 −1 2
−1 − 12 − 12 2
• accumulation leads to 9-point stencil

 
−1 −1 −1
 −1 8 −1 
 
−1 −1 −1
• exercise at home: compute Ab (k) and assemble global matrix

(and compare with stencil)
Typical Workflow
1. choose grid (essentially the Ω(k) ):
• quadratic or cubic cells
• triangles (structured, unstructured)
• tetrahedra, etc.
2. set up basis functions for each element Ω(k) ;
for example, at all nodes xi ∈ Ω(k )
ϕi (xi ) = 1
ϕi (xj ) = 0 for all j 6= i
3. for element stiffness matrix, compute all

Z
b (k) =
A ∇ϕj · ∇ϕi dx
ij
Ω(k)
4. accumulate global stiffness matrix A (and rhs)
Element-Oriented Computation on Unstructured
Meshes
Recall: FEM often used for complicated meshes:
(Re-)computation of element matrices required for each element?

• left grid: no (up to scaling with mesh size)
• right grid: yes?
⇒ requires a closer look at the computation of element matrices
Element Stiffness Matrices – Notation Revisited
• recall notation for local stiffness matrices A(k )
ij :
Z
(k)
Aij = ∇φj (x) · ∇φi (x) dx
Ω(k )
⇒ which basis functions φi , φj belong to element k?

• switch to element-local numbering of basis functions:
Z
(k)
Aµν =
b ∇φ(k) (k )
ν (x) · ∇φµ (x) dx
Ω(k)
(k) (k)
where φµ (x) = φi (x) and φν = φj (x) in Ω(k )
• requires a mapping φ(k) (k )
µ → φi (and φν → φj ),
i.e., a function ι(k) with i = ι(k) (µ) and j = ι(k) (ν)
Accumulation of Global Stiffness Matrix
• note the different dimensions of A(k) and A b (k) :
(k ) (k )
A omits zero rows and columns of A
b
• formulated via a projection matrix:

T (k ) 1 if i = ι(k) (µ)
A(k) = P (k) b (k ) P (k)
A with Pµ,i :=
0 otherwise
• recall computation of global stiffness matrix:
X X T (k ) (k )
A := A(k ) = P (k ) Ab P
k k
• hence, for each element:

b (k)
1. compute element stiffness matrix A
2. determine position of matrix elements Ab (k ) in A(k)
(k)
→ based on mapping ι (µ)
T (k ) (k)
3. add non-zero elements of A(k) = P (k) A b P to global matrix A
Element Stiffness Matrices and Meshes
Recall our initial question:
(Re-)computation of A b (k ) required for each element?

• left grid: no (up to scaling with mesh size)
• right grid: yes?
b (k) via reference elements
⇒ efficient computation of A
Simple Example: 1D Poisson
Consider FEM for 1D Poisson on an adaptive grid:
• grid points given as xk , not necessarily equidistant;
define elements Ω(k) = [xk , xk+1 ]
• piecewise linear basis functions (“hat functions”) φk (x);
φk (x) non-zero only on two elements: Ω(k−1) and Ω(k)
• mapping local to global indices: ι(k) (1) := k and ι(k) (2) := k + 1
• wanted: Ab (k) ∂ (k) (k)
φν (x) ∂ φµ (x) dx for every element k
R
µ,ν := (k) Ω ∂x ∂x
Introduce reference element Ωref with linear basis functions Φµ :
Φ1 (ξ) := ξ Φ2 (ξ) := 1 − ξ ξ ∈ Ωref := [0, 1]
and a mapping x = χ(k ) (ξ) for each Ω(k) defined as
χ(k ) (ξ) := xk + ξ(xk+1 − xk )

(k) (k )
such that φµ (x) = φµ χ(k) (ξ) := Φµ (ξ) for x ∈ Ω(k) .
Simple Example: 1D Poisson (2)
We need to compute
(k )
xk +1
Z χZ (1)
b (k) := ∂ (k) ∂ ∂ (k) ∂
A µ,ν φν (x) φ(k) (x) dx = φν (x) φ(k) (x) dx
∂x ∂x µ ∂x ∂x µ
xk χ(k ) (0)
g(b) Rb
f (g(t))g 0 (t) dt
R
Compare integration by substitution: f (x) dx =
g(a) a
Z1
b (k) = ∂ (k) (k) ∂ (k ) (k)
A µ,ν φ χ (ξ) φµ χ (ξ) (xk +1 − xk ) dξ.
∂x | ν {z } ∂x | {z }
0 =Φµ (ξ) =Φν (ξ)
Note that:
∂
• ∂ξ χk (ξ) = (xk +1 − xk ) corresponds to the factor g 0 (t) in the substitution
rule
• we still have to deal with the partial derivatives in x
∂f ∂f ∂g
We require the chain rule, ∂x = ∂g ∂x , to obtain:
∂ (k) (k) ∂ ∂ ∂ξ
φµ χ (ξ) = Φµ (ξ) = Φµ (ξ)
∂x ∂x ∂ξ ∂x
x − x k
As x = χ(k) (ξ) := xk + ξ(xk+1 − xk ), we have ξ = and thus:
xk+1 − xk
∂ ∂ξ ∂ 1
Φµ (ξ) = Φµ (ξ)
∂ξ ∂x ∂ξ xk+1 − xk
Finally, we get:
Z1
b (k) = ∂ (k ) (k) ∂ (k) (k )
A µ,ν φ χ (ξ) φ χ (ξ) (xk+1 − xk ) dξ
∂x ν ∂x µ
0
Z1
∂ ∂ 1
= Φν (ξ) Φµ (ξ) (xk+1 − xk ) dξ.
∂ξ ∂ξ (xk +1 − xk )2
0
Altogether, element stiffness matrices are computed as:
Z1
b (k) = 1 ∂ ∂
A µ,ν Φν (ξ) Φµ (ξ) dξ
xk+1 − xk ∂ξ ∂ξ
0
With Φ1 (ξ) = ξ and Φ2 (ξ) = 1 − ξ, we obtain that

b (k) = 1 1 −1
A
xk+1 − xk −1 1
Observations:
• the integral values only depend on the reference element Ωref = [0, 1] and
the reference basis functions Φµ , Φν
• all other element matrices are obtained by simple scaling
• we did not use that Φµ , Φν are linear functions;
same computation for more complicated basis functions!
Outlook: Extension to 2D and 3D
For 2D, 3D, etc., we require the multidimensional substitution rule:
Z Z
f (x, y, z) dΩ = f (x(ξ, η, ζ), y (ξ, η, ζ), z(ξ, η, ζ)) |J | dξ dη dζ
Ω(k ) Ωref
where |J | is the determinant of the Jacobian:

∂x ∂x ∂x

∂ξ ∂η ∂ζ
|J | = ∂y ∂y ∂y

∂ξ ∂η ∂ζ
∂z ∂z ∂z

∂ξ ∂η ∂ζ
and also the multidimensional chain rule.

Result:
• similar, element-wise computation of stiffness matrices via reference
elements
• |J | instead of factor x 1−x ; more complicated, in general
k +1 k
Outlook: Automatised Finite Element Assembly
Example: FeniCS Project, https://fenicsproject.org/ Numerical Programming 2
Compute a FEM-Solution of the 2D Poisson Equation:
from fenics import *

# Create mesh and define function space
mesh = UnitSquareMesh(8, 8)
V = FunctionSpace(mesh, ’P’, 1)
# Define boundary condition (omitted)
# Define variational problem

u = TrialFunction(V)
v = TestFunction(V)
f = Constant(-6.0)
a = dot(grad(u), grad(v))*dx
L = f*v*dx
# Compute solution Source: https://fenicsproject.org/pub/
u = Function(V) tutorial/html/._ftut1004.html#___sec19
solve(a == L, u, bc)
Outlook: Further Components and Aspects of FEM
Additional components and “features” of Finite Element Methods:
• mesh generation:
structured/unstructured meshes consisting of triangles, tetrahedra,
rectangles, cuboids, quadrilaterals, hexahedrals, prisms, . . .
• higher-order basis functions:
polynomials of higher degree in various dimensions; “nodal” vs. “modal”
basis functions, “spectral” elements, . . .
• quadrature rules for integration:
Gaussian quadrature, e.g., to compute integrals of weak forms;
resp. choice of quadrature points for certain elements?
• solving the system of equations:
e.g., set up a global system of equations explicitly, or work directly on
element representation?
. . . and many more . . .

Pde Fem

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Pde Fem

Uploaded by

Copyright:

Available Formats

Scientific Computing I

Module 7: An Introduction to Finite Element Methods

The Model Problem

FEM Main Ingredients

Example Problem: 1D Poisson

• 2D Poisson Equation on unit square:

• for general operator L instead of ∆: similar concept

→ allows additional weak solutions (esp. after integration by parts)

3. find weak solutions of simple form:

The weak form is obtained via 3 steps: −u 00 = f

2. integrate over the domain

⇒ “real solution” u also solves the weak form

• apply divergence theorem:

• choose functions v such that v = 0 on ∂Ω:

• Poisson equation with Dirichlet conditions:

• transformed into weak form:

• weaker requirements for a solution u:

FEM vs. Linear Algebra

the ϕj (x) are typically called shape or ansatz functions

• insert into weak formulation

• choose a basis {ψi } of a test space Vh ⊂ V

• aim: make matrix A sparse → most Aij = 0

Compute integrals of weak form:

• right hand sides (assume f (x) = α ∈ R):

• system of equations very similar to finite differences

• requires concepts for convergence of a sequence of functions

in terms of the norm induced by the bilinear form a (energy norm)

(details in lecture Numerical Programming II, e.g.)

Example: 1D Heat Equation

• apply discretisation in space using finite element method:

Mh : mass matrix, Ah : stiffness matrix, uh = uh (t)

Outlook: how to deal with inversion of mass matrix?

Element Stiffness Matrices

Reference Elements and Stiffness Matrices

Ω = Ω(1) ∪ Ω(2) ∪ · · · ∪ Ω(n)

• observation: basis functions are defined element-wisely

• and respective element systems:

• accumulate to obtain global system:

Some comments on notation:

• only 2 × 2 elements of A(k ) are non-zero

• consider case with only three unknowns (i.e., 3 basis functions):

• accumulation leads to 9-point stencil

• exercise at home: compute Ab (k) and assemble global matrix

3. for element stiffness matrix, compute all

4. accumulate global stiffness matrix A (and rhs)

(Re-)computation of element matrices required for each element?

⇒ which basis functions φi , φj belong to element k?

• hence, for each element:

(Re-)computation of A b (k ) required for each element?

Introduce reference element Ωref with linear basis functions Φµ :

Φ1 (ξ) := ξ Φ2 (ξ) := 1 − ξ ξ ∈ Ωref := [0, 1]

and a mapping x = χ(k ) (ξ) for each Ω(k) defined as

χ(k ) (ξ) := xk + ξ(xk+1 − xk )

With Φ1 (ξ) = ξ and Φ2 (ξ) = 1 − ξ, we obtain that

where |J | is the determinant of the Jacobian:

and also the multidimensional chain rule.

Compute a FEM-Solution of the 2D Poisson Equation:

from fenics import *

# Define variational problem

You might also like